Performance of aDeep-Learning Neural Network to Detect Intracranial Aneurysms from 3DTOF-MRA Compared to Human Readers.

Abstract

To study the clinical potential of adeep learning neural network (convolutional neural networks [CNN]) as asupportive tool for detection of intracranial aneurysms from 3D time-of-flight magnetic resonance angiography (TOF-MRA) by comparing the diagnostic performance to that of human readers. In this retrospective study apipeline for detection of intracranial aneurysms from clinical TOF-MRA was established based on the framework DeepMedic. Datasets of 85 consecutive patients served as ground truth and were used to train and evaluate the model. The ground truth without annotation was presented to two blinded human readers with different levels of experience in diagnostic neuroradiology (reader1: 2years, reader2: 12years). Diagnostic performance of human readers and the CNN was studied and compared using the χ2-test and Fishers' exact test. Ground truth consisted of 115 aneurysms with amean diameter of 7 mm (range: 2-37 mm). Aneurysms were categorized as small (S; <3 mm; N = 13), medium (M; 3-7 mm; N = 57), and large (L; >7 mm; N = 45) based on the diameter. No statistically significant differences in terms of overall sensitivity (OS) were observed between the CNN and both of the human readers (reader1 vs. CNN, P = 0.141; reader2 vs. CNN, P = 0.231). The OS of both human readers was improved by combination of each readers' individual detections with the detections of the CNN (reader1: 98% vs. 95%, P = 0.280; reader2: 97% vs. 94%, P = 0.333). ACNN is able to detect intracranial aneurysms from clinical TOF-MRA data with asensitivity comparable to that of expert radiologists and may have the potential to improve detection rates of incidental findings in aclinical setting.